Validation of Gated Blood-Pool SPECT Regional Left Ventricular Function Measurements

Patients

The study population consisted of 34 individuals (25 men and 9 women; mean age, 55 ± 13 y). All subjects were recruited from the inpatient and outpatient population at St. Francis Hospital and were in sinus rhythm at the time of both BP and cardiac MRI studies.

All of the subjects were being evaluated for various cardiac symptoms, including dyspnea on exertion and chest pain, and underwent gated SPECT BP studies and cardiac MRI. Twenty-four patients had a history of congestive heart failure, 17 had prior myocardial infarction, 29 had hypertension, and 8 had diabetes. Resting heart rates were 61 ± 13 beats per minute.

Normal limits for LV global and regional EF by BP AQ have been previously established (40 subjects, 20 men; age, 50 ± 16 y for all subjects) (7). For the cardiac MRI normal limits, 10 age- and sex-matched controls (5 men; age, 57 ± 8 y for all subjects) were selected from an established institutional database of normal global and regional cardiac MRI values. These values were based on cardiac MRI results obtained in healthy volunteers who had no history or evidence of cardiac disease on physical examination or electrocardiography (ECG) and who had normal results on rest and stress gated cardiac MRI myocardial perfusion studies.

This retrospective investigation was approved by the Institutional Review Board. All data were handled in compliance with the Health Insurance Portability and Accountability Act of 1996.

Cardiac MRI Data Acquisition

Cardiac MRI data were collected on a 1.5-T scanner (Sonata; Siemens Medical Solutions, Inc.). Breath-hold TrueFISP cine cardiac MRI was used (10), because this method produces images with a superior signal-to-noise ratio.

Multiplane localizers identified cardiac position and intrinsic LV long and short axes using standard iterative scouting techniques. Contiguous ECG-gated, breath-hold, segmented k-space TrueFISP images were obtained in the LV short-axis plane over the cardiac cycle with 128 × 256 matrices, using a field of view of 30–40 cm², depending on the patient's chest size. ECG gating was prospective. For the segmented technique, partial k-space data were acquired for each of the frames per heartbeat and then combined into a whole-frame image over several heartbeats. Image slice thickness was 8 mm, and temporal resolution was 40 ms or less, depending on the heart rate. Breath-hold duration was 10–16 s per image sequence. During 1 breath-hold, 2 contiguous planes were acquired. To minimize slice-position variations, data were always acquired during the patient's end expiration. Imaging extended from above the bifurcation of the main pulmonary artery to the apex of the heart over 11–13 tomographic sections (Fig. 1). Total data acquisition time was less than 30 min.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

MR images of 75-y-old male patient with history of congestive heart failure and hypertension with severe LV and RV dysfunction and abnormal LV EF (23%) at ED (A) and ES (B). Manually drawn epicardial and endocardial outlines are superimposed on images.

Cardiac MRI Calculations

Images were examined manually by 1 trained MRI analyst using commercially available software (MASS; Medis, Inc.). Each set of drawings was reviewed by 1 separate MRI expert. Slice cavity volume was determined as the product of slice thickness and cavity area, and ventricular volume was determined as the sum of slice cavity volumes. Cardiac MRI EF was computed from end-diastolic and end-systolic volumes. Using endocardial and epicardial manual drawings performed for 60° samples of short-axis cines over 11–13 tomographic slices, we computed regional cardiac MRI radial WM by MASS algorithms as the difference between end-diastolic and end-systolic endocardial radii, in millimeters. Cardiac MRI percentage wall thickening (WT) was computed as the percentage change in radial wall thickness from end-diastole (ED) to end-systole (ES). Cardiac MRI radial wall velocity (WV) was estimated as π*MW/(T_ED–ES), in units of centimeters per second, where T_ED–ES = time from ED to ES in units of seconds. This formula was derived by assuming that wall location at ED is the maximum of a cosine function for which wall location at ES is a minimum and taking the derivative of the cosine to obtain WV. All 66–78 cardiac MRI regional values then were resampled into segments corresponding to the conventional LV 17-segment model by weighted averaging, with weights determined by the percentage overlap of surface areas of the 17 segments as projected onto surface areas of the original 66–78 locations.

BP Image Acquisition

^99mTc-pertechnetate (925 MBq [25 mCi]) was injected 30 min after an injection of 5 mg of stannous pyrophosphate. Data were acquired using dual-detector γ-cameras (Forte or Vertex; ADAC Inc.) equipped with low-energy, high-resolution collimators. Data were collected at 64 projections over a 180° circular arc and 64 × 64 tomograms (pixel size, 4.6 mm), acquired for 20 s/projection for 15–16 min. Tomograms were acquired at 16 frames per R-R interval, using an R-wave window of ±20% of mean preacquisition heart rate (11).

Spatial smoothing before reconstruction was performed using Butterworth filtering (cutoff, 0.35 cycles/cm; power, 5.0) for gated tomograms, followed by ramp filtering in the transaxial plane. Filtered images were backprojected to form transaxial slices, which were reoriented into short-axis sections by manually choosing approximate LV symmetry axes, using standard software. Reconstruction limiting regions and angles were chosen to ensure that the entirety of both ventricles, as well as the atria and the main pulmonary artery, were included in the field of view of reoriented images. Input tomograms to BP software were reoriented manually to achieve optimal separation of right ventricular (RV) and LV activity seen in tomographic short-axis views.

BP Visual Assessment

Independent visual assessment of LV global and regional WM was performed by 1 experienced nuclear cardiologist, without knowledge of BP AQ or cardiac MRI calculations or any other clinical test results. BP visual assessment was performed with cines displayed simultaneously in orthogonal views without superimposed computed outlines (Fig. 2). LV WM for each of the 17 segments was graded on a 5-point scale (0, normal; 1, mildly hypokinetic; 2, moderately hypokinetic; 3, akinetic; and 4, dyskinetic). On the basis of a segment judged to be at least moderately hypokinetic, BP regional visual assessment readings also were dichotomized.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Screens of cines used to visually analyze BP-gated SPECT regional WM abnormalities for same patient in Figure 1 at ED (A) and ES (B).

On a separate occasion, the same observer, unaware of previous assessments and all other measurements, repeated visual assessments so that there were 2 sets of visual readings (initial visual assessment of BP-gated SPECT WM and second independent visual assessment of BP-gated SPECT WM). Reading sessions were separated by more than 1 mo to avoid recall bias.

BP Image Processing

Previously described BP algorithms (8) were modified to include new regional function subroutines written in interactive data language (RSI, Inc.) and executed on standard desktop personal computers running Windows operating systems (Microsoft Corp.) (7). Maximum-count sites in likely ventricular volumes identified LV mid-chamber locations. Valve planes were generated by a combination of shape information incorporated with phase information. Volumes were computed from the number of voxels greater than 50% of maximum LV volumetric counts, and EF was computed from count changes in areas with counts greater than 35% of maximum LV volumetric counts.

For each segment of the R-R interval, RV and LV volumes were divided into 17 subvolumes. The counts of each subvolume were fit to third-order Fourier harmonics, using a moving frame of reference (12). Each of the 17 LV segments corresponded to the conventional LV mapping (Fig. 3) (13). Regional EFs were calculated from the count changes from ED to ES for each of the 17 subvolumes.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

Polar map of BP-gated SPECT LV regional EF (percentage) for patient in Figure 1.

All BP data were processed by AQ, with no operator interventions. Results of all numeric calculations were exported by the algorithms automatically to Excel (Microsoft) spreadsheets for subsequent statistical analysis.

Statistics

Statistical analyses were performed using commercially available software (Medcalc, version 7.5.0.0.; Medcalc Software, Inc.). Values are reported as mean ± SD. Frequencies and percentages were used to characterize categoric variables. One-way ANOVA was used to determine whether variations were significant among the 17 LV territories for WM, WT, or WV values for controls. Paired t tests were used to test the hypothesis that mean values were not different between BP AQ and cardiac MRI computations. Correlation between continuous variables was determined using linear regression, which generated Pearson correlation coefficients (r). Bland–Altman analyses of measurement differences plotted versus mean values were used to assess biases, trends, and limits of agreement. Using dichotomous cardiac MRI WM, WT, and WV abnormality scores as the reference standards, we performed receiver-operating-characteristic analyses of BP AQ and BP visual assessment to correctly identify which LV segments were abnormal for all LV segments and for segments in the left circumflex (LCX), left anterior descending (LAD), and right coronary artery (RCA) territories. Agreement between methods for identifying which LV segments were abnormal was determined using the κ-statistic (14). The McNemar test was used to determine whether differences were significant for paired proportions of normal versus abnormal findings for BP AQ versus cardiac MRI in identifying patients with global EF less than 35% (15) and patients with EF less than 50%. The McNemar test also was used to determine whether differences were significant for categorizing LV segments as abnormal by BP AQ and BP visual assessment versus cardiac MRI, for all segments and for segments subgrouped by coronary territory. A P value of less than 0.05 was considered to be statistically significant. All statistical tests were 2-tailed.

Global Parameters

Algorithms were run for BP data for all 34 subjects. BP global and regional calculations took less than 1 min per patient. Manual endocardial and epicardial drawings for cardiac MRI global and regional calculations required about 1 h per patient.

Cardiac MRI global EF was significantly different between controls and study subjects (58% ± 7% vs. 46% ± 18%, P = 0.04), as was end-systolic volume (59 ± 22 mL vs. 121 ± 81 mL, P = 0.02). LV global EF ranged from 12% to 75% for AQ and from 14% to 75% for cardiac MRI. AQ EF was similar to cardiac MRI EF (47% ± 18% vs. 46% ± 18%, P = 0.30). Correlation was significant for AQ versus cardiac MRI EF (P < 0.0001) (Fig. 4). Bland–Altman analysis showed that EF differences did not correlate with EF means (P = 0.64), with no bias (P = 0.95) or trend (P = 0.64), and with limits of agreement of −9.8% to 11.7%. End-diastolic and end-systolic volumes ranged from 28 to 499 mL for AQ and from 35 to 525 mL for cardiac MRI. Volumes were similar for BP and cardiac MRI (152 ± 86 mL vs. 160 ± 90 mL, P = 0.11) and correlated as r = 0.87 (P < 0.0001). Bland–Altman volume differences were uncorrelated versus mean values (P = 0.45), with no bias (P = 0.93) or trend (P = 0.45) and limits of agreement of −97 to 80 mL. Agreement between BP AQ and cardiac MRI was “very good” (14), with no significant disagreements for identifying the patients with EF less than 50% and less than 35% (Table 1) (15).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

BP measurement of global LV EF vs. cardiac MRI–quantified global LV EF. CMR = cardiac MRI.

Regional LV Parameters

Cardiac MRI regional values were different between controls and patients for WM (6.5 ± 1.7 mm vs. 5.0 ± 2.8 mm, P < 0.0001), WT (53% ± 26% vs. 33% ± 30%, P < 0.0001), and WV (3.4 ± 1.3 cm/s vs. 2.6 ± 2.8 cm/s, P < 0.0001). Variations among cardiac MRI values were significant (ANOVA, P < 0.05) for WM, WT, and WV when all 17 segments were included and when only the 3 septalmost territories were examined, but variations were not significant when the 3 septalmost territories were excluded. These variations reflect the methodologic difficulties involved in drawing cardiac MRI LV contours in the vicinity of the LV outflow tract. Therefore, the 3 septalmost segments were excluded from all subsequent analyses comparing cardiac MRI, BP AQ, and visual assessment BP measurements. Cardiac MRI lower limits of normal were 3.2 mm for WM; on that basis, there were 141 segments with abnormal WM and 335 segments with normal WM.

Receiver-operating-characteristic analysis showed that accuracy of BP AQ for predicting abnormal WM was greater than the accuracy of BP visual readings (P = 0.0001) (Table 2; Fig. 5A). AQ was significantly more accurate for LCX than for LAD or RCA territories (P = 0.0002) and significantly more accurate than visual readings for LCX and RCA territories (P = 0.01–0.002) (Table 2; Fig. 5). Accuracy of AQ to predict abnormal WM was not influenced by the severity of LV dysfunction. The accuracy for all territories for all subjects (88% ± 2%) was similar to the accuracy for the subgroup with global EF less than 50% (89% ± 2%, P = 0.78) and similar to the accuracy for the subgroup with global EF less than 35% (87% ± 2%, P = 0.78). AQ was more accurate for predicting abnormal WM than for predicting abnormal WT (P = 0.002) and was more accurate than visual assessment for predicting abnormal WT for the LCX territory (Table 2). AQ for predicting abnormal WV was higher than the accuracy of visual readings, with highest accuracy versus WV for LCX territories (Table 2).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5. 

Receiver-operating-characteristic curves to predict LV segments with abnormal WM as seen on cardiac MRI, for AQ BP-gated SPECT EF, initial visual assessment (V1), and second visual assessment (V2) for all territories (A) and for LCX (B), LAD (C), and RCA (D) territories only.

Agreement with cardiac MRI was “moderate to good” for BP AQ for identifying segments with abnormal WM, with no significant disagreements. Agreement was only “moderate” for visual assessment, with significant RCA disagreements (Table 3) (14).

McNemar differences for identifying abnormal segments were not significant for AQ versus WM (P = 0.14) or AQ versus WV (P = 0.30) but were significant for WT versus WM (P < 0.0001), AQ versus WT (P < 0.0001), AQ versus visual assessment (P = 0.001), initial visual assessment versus second visual assessment (P = 0.003), visual assessment versus WM (P < 0.0001), and visual assessment versus WV (P = 0.0001). For detecting paradoxical apical WM, explicitly, AQ disagreements from WM were not significant (McNemar difference, 3%, P = 1.0).

Measurement Reproducibility

The BP global and regional calculations of this investigation were totally automated; consequently, these calculations were perfectly reproducible. Interobserver agreement of repeated BP visual assessment readings for all territories was “moderate to good,” with no significant differences (Table 4) (14).